Affirmative Action, Education and Gender: Evidence from
India.∗
Guilhem Cassan†
January 30, 2017
Abstract: This paper studies the impact of affirmative action policies on educational attainment in India. Using plausible exogenous variation, they are shown to have
widely differing impact across genders, with males gaining about 2 years of education
while females not being affected and maybe losing. This underlines the fact that groups
cumulating discrimination (caste and gender) may not be sufficiently well protected by
“simple” affirmative action policies.
JEL Classification: I24; O15; H41
Keywords: identity; scheduled caste; quota; affirmative action; gender; India; education; intersectionality.
∗
I am grateful to Gani Aldashev, Denis Cogneau, Christelle Dumas, Esther Duflo, Hemanshu Kumar,
Eliana La Ferrara, Sylvie Lambert, Andreas Madestam, Marco Manacorda, Annamaria Milazzo, Barbara
Petrongolo, Rohini Somanathan, Vincenzo Verardi and Yanos Zylberberg, as well as participants in the
seminar and conferences in which this paper has been presented. Marion Leturcq’s and Lore Vandewalle’s
comments on an earlier draft vastly improved this paper. I am indebted to Julien Grenet for his help in the
data collection, and to Ashwini Natraj, with whom I started this project. This paper is produced as part
of the project “Actors, Markets, and Institutions in Developing Countries: A micro-empirical approach”
(AMID), a Marie Curie Initial Training Network (ITN) funded by the European Commission. Funding
from the CEPREMAP is gratefully acknowledged. This research used resources of the “Plateforme
Technologique de Calcul Intensif (PTCI)” (http://www.ptci.unamur.be) located at the University of
Namur, Belgium, which is supported by the F.R.S.-FNRS under the convention No. 2.4520.11. The
PTCI is member of the “Consortium des Equipements de Calcul Intensif (CECI)” (http://www.cecihpc.be). All remaining errors are mine.
†
University of Namur, CRED. Email: [email protected].
1
Introduction
Evaluating affirmative action policies is by construction a difficult exercise. As a result,
while these policies are often widely debated, the debate is rarely well informed. India, in particular, has implemented the largest affirmative action program in the world,
which targets the low castes and in particular the “Scheduled Castes” (SC)1 . Those
policies have been controversial since they were systematically introduced after the Independence. In the early 90s, with the enlargement of affirmative action policies to a
large new group of relatively well off castes (the “Other Backward Classes”, or OBC)
the debate has become particularly intense, as higher castes began to fear competition.
The debate has notably focused on quotas in higher education, in which competition
with high caste was the strongest. Following this trend, most of the research has dealt
with the consequences of quotas in universities (Bertrand et al., 2010; Krishna and Frisancho Robles, 2012) or of quotas for OBC (Khanna, 2013). However, according to the
2011 Census, only 56% of the SC population aged 20 and above was literate (against
66% in the whole population), only 30% had attained an education level higher than
primary (against 41%), and only 6% had gone beyond secondary schooling (against
11%). And these vary broadly by gender: only 43% of women aged 20 and above are
literate against 66% for the whole population, 21% went above primary school and 3.8%
went beyond secondary schooling2 . Hence, the current focus on affirmative action in
higher education entirely neglects the most weaker segment of the Indian population,
the Scheduled Castes which in their immense majority never even reach university. Focusing on university level affirmative action programs is not sufficient to get a broad
picture of the effect of affirmative action policies targeted to low castes in general and
to SC in particular. As a result, the diagnostic drawn several decades ago by Chalam
(1990), Chitnis (1972) or Galanter (1984) still holds: we have very little knowledge about
the impact of these policies on SC. This paper ambitions to contribute to filling this gap.
I show in this paper that the overall effect of affirmative action policies on educational attainment of SC is quite small, if any. However, this small overall effect hides
large heterogenous effects across genders. Indeed, males gain up to 2.5 years of education
while females are not affected and may even have suffered from the access to the SC
1
The “Scheduled Castes” are the castes traditionally the most discriminated against. They represent
approximately 17% of the population in 2011.
2
And this diagnostic remains true for younger cohorts, as focusing on the 25-29 cohort yields better
but still very low figures: 70% literacy (60% for female), 42% (resp. 34%) above primary school and 8%
(resp. 6.5%) above secondary school.
2
status. This result underlines the fact that populations that cumulate disadvantages,
that are at the intersection between different discriminated groups, such as women of
low castes, may not be sufficiently protected by policies which do not take into account
this cumulative discrimination. Strikingly, the gains for the males come from the first
years of education: affirmative action policies seem to push families to put their children
in the schooling system, in primary school in particular, while there does not seem to be
any robust increase in later years of education. Hence, this paper adds two important
contributions to the literature on affirmative action in India. First, by focusing only on
the higher end of education, impact evaluations may miss important effect of affirmative
action policies. Second, by putting the accent on the gender difference of the impact of
affirmative action, this paper emphasizes the heterogenous effect of affirmative action
policies within the treated population. As such, it relates to the literature on gender
discrimination (Jensen, 2012) and in particular to the literature on asymmetric effect
of social policies in developing countries (Foster and Rosenzweig, 2003; Rosenzweig and
Schultz, 1982; Ashraf et al., 2015) as well as to the “intersectionality” literature in political science, which studies the consequences of the accumulation of discriminations
(Hughes, 2011, 2014; Jensenius, 2015).
The research design of this paper relies on a unique natural experiment on the access
to the SC status. As a matter of fact, the list of the castes considered as SC were drawn
by each Indian state at the Independence. In 1956, the borders of the Indian states have
been redrawn, while the lists of SC remained unchanged. This created many within
State discrepancies: the SC status of the members of a same caste could vary across
space3 . This situation lasted until 1976, when the list of castes considered as SC were
harmonized within each state, allowing 2.4 million individuals (Government of India, ed,
1978) to have access to the SC status4 . This historical setting allows me to assess the
fate of the individuals who had access to the SC status from 1976.
As the treatment status varies across castes, and within caste across time and space,
this paper innovates by coding the treatment status of individuals based on their precise
3
In practice, each state had several lists of SC, by regions usually grouping several districts. Those
regions are listed in Appendix A.
4
Note that this means that the identification does not rely on policy variation across states, but on
variation to exposure to affirmative action policies within a single state. This specific setting allows to
safely attribute any change observed to only a change in access to the Scheduled Caste status and not
to any other state wide policy.
3
caste name5 instead of their declared beneficiary status6 . By using the precise caste
name I can use demanding specifications, which identify the effects of the policy within
a single caste, something that, to the best of my knowledge, has never been done in
the literature. As a matter of fact, the main specifications rely on caste*cohorts and
region*cohorts fixed effects, which allow to identify the treatment effect using only the
within caste-cohort and within region-cohort variation. In a robustness check, I show
that adding a specific trend for the treatment group does not alter the results. Hence,
the results presented in this paper are not due to pre-existing trends, and can safely be
attributed to the acquisition of the Scheduled Caste status.
In the first section of this paper, I will present the context and the natural experiment
exploited in this paper. I will then describe the data as well as the empirical strategy,
which will open the way to the presentation of the results, their discussion and various
robustness checks (varying treatment across cohorts, differential trends, migration and
identity manipulation).
1
Context
1.1
Affirmative action in India
While the first affirmative action policies for Scheduled Castes were implemented under
British rule, it is not before the Independence that a systematic positive discrimination
policy was implemented. Also called “reservations”, it has 3 main dimensions: political
representation, education and public employment. Electoral quotas are indeed implemented for SC in all federal and state assemblies, as well as in panchayat elections (since
the mid 90s), while public recruitment is subject to a quota policy, with a share of new
openings to be reserved for low caste groups. Finally, in education, affirmative action
consists in various policies. There are quotas in higher education institutions, free secondary schooling, as well as access to scholarships, specific schools and hostels and free
mid day meals7 . Positive discrimination might thus affect schooling through various
channels. By reducing the cost of education, it favors longer studies in the cost-benefit
arbitrage of the household, while quotas in higher education will help the pursuit of
5
Their “jati”, that is the endogamous group which is the relevant social category in every day life, as
opposed to their caste category, such as “Scheduled Caste”, which is an administrative category.
6
Indeed, most of the micro level literature on affirmative action in India uses the the household’s
declaration of its SC status as a basis for identifying the “treatment” group (Khanna, 2013; Hnatkovska
et al., 2012, 2013; Prakash, 2009)
7
See Galanter (1984), chapter 3.B for a presentation of the main educational policies targeting low
castes apart from quotas in higher education.
4
studies after secondary schooling. Moreover, the quotas in public employment increases
the returns to education. Hence, this paper does not evaluate the effect of affirmative
action in the educational sector but the effect of the package of affirmative action policies
on educational attainment.
1.2
The definition of the Scheduled Castes
How are the beneficiaries of these affirmative action policies determined? As underlined
by Galanter (1984), one of the main difficulties of the creation of a list of Scheduled Castes
is the criteria of inclusion in that list. Indeed while “untouchability” is the prerequisite
to be considered a SC, defining “untouchability” is actually not straightforward since its
meaning varies across the subcontinent. Hence, the Constitution of 1950 avoids to define
a clear concept of untouchability and only provides a procedure of designation8 that each
State is to follow. This allowed for the possibility of inconsistencies across States9 as well
as within States, as certain States decided to give the SC status to certain castes only in
certain areas. Despite those inconsistencies, the lists were revised only four times since
the Independence10 . With an increase of 2.4 million SC over an original population of
80 million SC, the Scheduled Castes and Scheduled Tribes (Amendment) Act of 1976
was the most dramatic change in the list of SC in India.
1.3
The Scheduled Castes and Scheduled Tribes Orders (Amendment)
Act of 1976
In 1956, India reorganized the borders of its States along linguistic lines11 . But, as the
borders of the States were redefined, the State-wise SC lists remained unchanged. This
led to a large increase of within state discrepancies in the SC lists, since there were
now systematically several SC lists per state. Hence, from 1956, the number of castes
considered as SC in one part of a State and not in an other part of the same State vastly
increased. It is only in 1976 that the SC lists were harmonized within states by the SC
8
“castes, races or tribes or parts of or groups within castes, races and tribes which shall for purposes
of this Constitution be deemed to be Scheduled Castes in relation to that State.”
9
Bayly (1999) gives the example of the Khatik caste, considered as SC in Punjab, but classed as a
“forward” caste in Uttar Pradesh, a neighboring State at the time of the establishment of the lists.
10
The first change, in 1951 was a matter of correcting small anomalies. The second change, in 1956
mainly affected Rajasthan and Uttar Pradesh, and also allowed all Sikh untouchable castes to claim SC
status, while the fourth change of 1990 allowed the Buddhists to have access to the SC status in all the
States. The most important change took place in 1976 and is described in more details in the paper.
11
And in 1960, the states of Maharashtra and Gujarat were created from the former state of Bombay.
5
and ST (Amendment) Act of 1976, also called the “Area Restriction Removal Act”12 .
This Act removed almost all intra-State restrictions13 . This removal of restrictions led
to an increase of 2.4 million of the SC population14 . The list of these areas of restriction
(which I’ll call “regions” in the paper) is in Appendix A. In practice, the paper uses 31
regions.
Hence, individuals from the same caste could be considered as SC in one part of
a State but not in an other until 1976, when anyone whose caste was on the SC list
somewhere in a state would be considered SC in that state. This situation thus creates
a natural experiment setting in which the members of a same caste faced different access
to affirmative action status within the same state. This unique historical event creates a
plausible exogenous variation in the SC status across individuals, allowing me to assess
the causal impact of the SC status on educational attainment.
2
Data and Empirical Strategy
2.1
Data and descriptive statistics
The National Family and Health Survey of 1998-99 (NFHS2) is, to my knowledge, the
only dataset offering both the precise caste name (the “jati”) of respondents, their district of residence and a sufficient sample size to perform the type of analysis done in this
paper15 .
12
The reason for the list not to be adjusted to the new borders was the slowness of the administration:
“It has been mentioned in the last report that the President has issued the SC and ST Lists (Modification) Order, 1956, specifying the SC and ST in the re-organized States. As these lists had to be issued
urgently for the re-organized States, it was not possible to prepare comprehensive and consolidated lists
and therefore, the SC and ST had to be specified in these list territory-wise within each re-organized
State” (Government of India, ed, 1958). But not only did the administration fail to change the lists on
time, it failed to do so for a period of twenty years. The yearly reports of the Commissioner on SC and
ST are particularly telling in this aspect, as many of its yearly occurrences refer to the fact that “[...]the
question of preparation of comprehensive lists of SC and ST for the reorganized States [...] remained
pending [...]” (Government of India, ed, 1960).
13
According the Galanter (1984) their number dropped from 1,126 to 64.
14
The states affected by the reform were Andhra Pradesh, Bihar, Gujarat, Karnataka, Kerala, Madhya
Pradesh, Maharashtra, Rajasthan, Uttar Pradesh, Tamil Nadu, West Bengal and Himachal Pradesh.
The analysis excludes Himachal Pradesh, as when this Union Territory was granted the State status
in 1966, large portions of the state of Punjab were also transferred to it: hence, between 1956 and
1966, the population living in the contemporary borders of Himachal Pradesh were exposed to different
State policies, preventing me from identifying the sole effect of the access to the SC status. Results are
unaffected by the inclusion of Himachal Pradesh. Note that if the changes in border mainly affected the
South of India, the Area Restriction Removal Act also affected northern states such as Bihar and Uttar
Pradesh, who had within states variations in their SC list before 1956.
15
For example, the 2004-5 round of the NFHS does not contain information on the district of residence
of the respondents.
6
The jati names present in the NFHS2 are present only in a raw form. As a consequence, an effort has been made to clean the names of the jatis of Scheduled Castes in a
systematic manner. The procedure followed was to send the list of the raw caste names
with the list of castes and synonyms of Kitts (1885)16 as well as the a second caste list
which (supposedly) had been used for the coding of the first round of the NFHS17 to a
data entry team based in India for them to match names from the NFHS2 to the two
lists. I then refined this matching using the state wise list of Scheduled Caste synonyms
available in the Census of 1971 and 1981. This methodology has led to the identification
of a total of 179 SC jati, defined at the state level18 . Using the cleaned jati names, the
1971 and 1981 Census lists of Scheduled Castes and the district of residence of households, I am able to identify the households that were granted the SC status in 197619 .
Throughout this paper, I will call this population “New SC”, as opposed to the “Old
SC” who were already on the list in 1976. This methodology differs from most studies
of caste using nationally representative surveys. Indeed, while these studies generally
rely on the declarations of the respondent on their Scheduled Caste status, I use the
jati name of the respondent to attribute the SC status, which allows to have within jati
estimates of the access to the SC status.
Tables 1 and 2 provide the summary statistics for the variables used throughout the
paper. I restrict the sample to individuals born between 1951 (below school age in 1956,
year of the redefinition of borders) and aged 17 and above at the time of the survey. It
can be seen that the educational attainment in the sample is lower for females than for
males, and that overall, “Old SC” have a higher level of education than “New SC”. It
can also be noted that the declared religions are somewhat different across groups. All
the results shown are robust to restricting the sample to only Hindus20 .
16
This source was used at it had been the base of the coding of caste names in the first round of NFHS
(1992-3).
17
Indeed, the first round of the NFHS (1992-3) also offers jati and district identifiers. This project
initially planned to use both first and second NFHS rounds. However, the first round turned out to be
unusable as the documentation on the codes of caste (which is based on Kitts (1885)) does not correspond
to the codes present in the data. It has also come to my knowledge that a different list of caste codes
had been used for the first round of NFHS, partially overlapping Kitts (1885), and which I managed to
acquire. However, this second list appeared to be unknown to the central organization of NFHS and
also to not correspond perfectly to the codes present in the raw data either. As a consequence, the first
round of the NFHS could not be used.
18
There are two main reasons to define the jatis at the state level. First, the same jati name might
not mean the same thing in different areas of India. Second, since the list of synonyms of SC present in
the 1971 and 1981 Census is state based, the coding of the synonyms was also state based.
19
In order to do so, I simply match the 1999 districts of the NFHS2 data to their 1971 counterparts
7
Table 1: Descriptive Statistics: Male population.
New SC
Mean
Std. Dev.
Old SC
Mean
Std. Dev.
Education variables
Years of Schooling
Literacy
Schooling
Primary Completion
Incomplete Secondary
Secondary Completion
Higher Education Completion
Control variables
Urban
Hindu
Christian
Sikh
Buddhist/Neo Buddhist
No Religion
Declares to be SC
5.01
0.61
0.64
0.51
0.45
0.20
0.13
4.82
0.49
0.48
0.50
0.50
0.40
0.34
5.97
0.68
0.72
0.61
0.53
0.26
0.16
4.84
0.47
0.45
0.49
0.50
0.44
0.37
0.24
0.99
0.01
0.00
0.00
0.00
0.68
0.43
0.11
0.09
0.00
0.00
0.04
0.47
0.30
0.91
0.03
0.00
0.06
0.00
0.82
0.46
0.29
0.17
0.05
0.23
0.01
0.38
N
680
6640
Table 2: Descriptive Statistics: Female population.
New SC
Mean
Std. Dev.
Education variables
Years of Schooling
Literacy
Schooling
Primary Completion
Incomplete Secondary
Secondary Completion
Higher Education Completion
Individual and household variables
Urban
Hindu
Christian
Sikh
Buddhist/Neo Buddhist
No Religion
Declares to be SC
N
2.11
0.26
0.28
0.21
0.19
0.08
0.05
4.00
0.44
0.45
0.41
0.39
0.28
0.23
2.87
0.35
0.39
0.30
0.25
0.11
0.07
4.25
0.48
0.49
0.46
0.43
0.31
0.25
0.20
0.98
0.01
0.00
0.00
0.00
0.67
0.40
0.12
0.11
0.00
0.00
0.04
0.47
0.29
0.90
0.03
0.00
0.06
0.00
0.82
0.45
0.29
0.17
0.05
0.24
0.00
0.38
653
8
Old SC
Mean
Std. Dev.
6636
2.2
Identification Strategy
As the SC are a different population from the general population, much poorer in particular, comparing the “new” SC to the general population does not allow to identify
the specific effect of the access to the SC status on educational attainment from other
social policies.
Hence, the natural counterfactual are the SC already on the lists in 1976. Due the
specific setting of the natural experiment, I am able to compare within the same caste,
within the same state, the fate of those who had access to the SC status in 1976 against
those who had access to it earlier. This calls for a difference in differences specification,
the first difference comparing cohorts too old to benefit from the access to the SC status
in 197721 in terms of education to those that were young enough. Throughout the paper,
I will consider as “young enough” the cohorts aged either 6 or 11 years old and below in
1977. The second difference would then compare those who had access to the SC status
prior to 1977 to those that did not.
I will thus run regressions of the type:
Eduidt = constant + βN ewSCid + δN ewSCid ∗ posttreatmentt + γposttreatmentt
+ λXidt + idt
(1)
Where Eduidt is a measure of educational attainment of individual i born in year t
and residing in region d, N ewSCid a dummy indicating whether individual i residing in
region d is member of a caste added to the SC list in 1976, posttreatmenti a dummy
taking value 1 if individual i is in a treated cohort and Xidt a set of control variables.
Note that since the treatment is attributed at the caste and region level, this indicates
the need to two way cluster the standard errors along those two dimensions. As there
are 31 regions, I check the “robustness” of the standard errors by two-way clustering and
bootstrapping at the region cluster level using the method proposed by Cuskey (2015).
All tables present simple two-way clustered standard errors as well as 95% confidence
intervals calculated with Cuskey (2015), which do not alter the statistical significance of
the results.
using the Indian Administrative Atlas of 2001, which follows district changes over time.
20
Results available on request.
21
Year of implementation of the 1976 change.
9
3
Results
3.1
Main specifications
The identification strategy of this paper relies on within caste variation in the exposure to
the SC status. The specifications use caste-5 years cohort FE and region-5 years cohort
FE. That is, the estimation is solely identified on within region-cohort and within castecohort variation. However, the first cohort to be exposed to the treatment is not well
defined a priori. Indeed, treated individuals are the ones that would still be at school
at the time of the reform. Since I do not have that information (I observe only the
schooling choices after the reform has been implemented, which is endogenous) I resort
to a rule of thumb. I assume that the first treatment cohort would be the cohort that
reaches the median age of schooling in 1977, that is cohort 1966 for males22 and cohort
1971 for females23 . I then repeat the regression twice, first with 1966 as the cut off years
and second with 1971.
Tables 3 presents the results of those specifications. Access to the SC status does
not seem to affect the level of education much. However, once the population is divided
by gender, very different effects appear. Indeed, it can be seen that the SC status leads
to an increase of 1 to 2.1 years of schooling for males, spread across pre secondary
schooling levels. For women, access to the SC status leads to a decrease of 0.7 to 1 year
of schooling, driven by a decrease in primary school completion. Note that this is a
relative decline: the context is one of overall increase of schooling, so the treated group
does not see its absolute level of education decrease. Note also that the treatment effect
seems larger for males when the treatment year is 1966 instead of 1971, which points to
the fact that male cohorts were treated earlier than 1971, while the contrary is true for
females. I will test more formally that insight in the next subsection.
Hence, the overall null effect of the access to the SC status on educational attainment
hides asymmetrical effects by gender which cancel each other out in the aggregate.
Indeed, while males do benefit from the SC status, it seems that females do not, and
may actually lose from the status.
3.2
Varying treatment year
We have seen that male’s treatment effect seemed to be higher when the treated cohort
was assumed to be 1971 instead of 1966, and that the contrary was true for female. This
22
23
11 years old in 1977 for a median of 5 years of schooling for the cohorts born up to 1971.
6 years old in 1977, for a median of 0 years of schooling for the cohorts born up to 1971.
10
11
7289
-0.09
[0.077]
[-0.21, 0.04]
-0.113
[0.083]
[0,25, 0.04]
7320
0.077
[0.054]
[-0.02, 0.17]
0.13
[0.083]
[-0.005, 0.26]
14609
-0.012
[0.065]
[-0.12, 0.08]
0.006
[0.057]
[-0.1, 0.11]
Schooling
7289
-0.114**
[0.053]
[-0.2, -0.03]
-0.111**
[0.050]
[-0.19, -0.03]
7320
0.146**
[0.060]
[0.05, 0.24]
0.261***
[0.075]
[0.13, 0.39]
14609
0.025
[0.053]
[-0.06, 0.11]
0.077
[0.050]
[-0.01, 0.16]
Primary
Some Secondary
7289
-0.081
[0.066]
[-0.19, 0.02]
-0.071
[0.060]
[-0.17, 0.03]
7289
-0.095*
[0.051]
[-0.18, -0.01]
-0.035
[0.034]
[-0.09, 0.02]
7320
0.119**
[0.050]
[0.03, 0.2]
0.227***
[0.057]
[0.12, 0.33]
Female Sample
7320
0.158**
[0.063]
[0.05, 0.26]
0.206**
[0.090]
[0.06, 0.35]
14609
0.034
[0.038]
[-0.3, 0.09]
Male Sample
14609
0.016
[0.063]
[-0.08, 0.11]
Full Sample
0.056
0.113***
[0.050]
[0.040]
[-0.02, 0.13]
[0.04, 0.18]
Literate
7289
-0.016
[0.050]
[-0.1, 0.06]
0
[0.035]
[-0.05, 0.05]
7320
0.002
[0.097]
[-0.15, 0.14]
0.086
[0.068]
[-0.02, 0.19]
14609
0.006
[0.055]
[-0.08, 0.09]
0.056
[0.041]
[-0.01, 0.12]
Secondary
7289
-0.043
[0.046]
[-0.11, 0.03]
-0.002
[0.034]
[-0.05, 0.05]
7320
-0.08
[0.075]
[-0.20, 0.03]
0.011
[0.062]
[-0.07, 0.10]
14609
-0.032
[0.050]
[-0.11, 0.45]
0.024
[0.041]
[-0.04 ,0.09]
Higher Education
*Significant at the 10%, **significant at the 5%, ***significant at the 1%. Standard errors two way clustered at the jati and region level in parenthesis. 95%
confidence interval from two way clustering at the jati and region level, bootstrapped at the region level in brackets. Controls include regions FE, state-cohort
FE, jati FE, gender FE and religion of household head FE. Population born after 1950 and aged 17 and above at the time of the survey.
7289
-1.003
[0.619]
[-2.10, -0.05]
New SC*post 1971
N
-0.684
[0.443]
[-1.44, -0.02]
7320
0.968*
[0.518]
[0.07, 1.84]
2.164***
[0.663]
[1.09, 3.09]
New SC*post 1966
N
New SC*post 1971
New SC*post 1966
14609
0.061
[0.526]
[-0.85, 0.83]
New SC*post 1971
N
0.831*
[0.450]
[0.09, 1.48]
New SC*post 1966
Years of Schooling
Table 3: Educational attainment. OLS regressions.
has a logical interpretation: given the education differential between boys and girls,
much less women were still at school at age 11 in 1977 compared to males. As a result,
the cohort of treatment should vary across gender. To test more formally this intuition,
I will allow the treatment effect to vary both for cohorts born in 1966 and after and for
cohorts born in 1971 and after, running the following regression:
Eduidt = constant + βN ewSCid + δ1 N ewSCid ∗ post1966t + γ1 post1966t
+ δ2 N ewSCid ∗ post1971t + γ2 post1971t + λXidt + idt
(2)
This will allow me to test if indeed males are treated in earlier cohorts that females. Table
6 presents the results of such a specification. It can be seen that while the female result
loses significance, the size of the coefficient is larger for 1971 than for 1966, suggesting
that if anything, women from later cohorts are more affected than older cohort. For
males, it is clear that the affected cohort start after from 1966 and not 1971. This
reinforces our conclusion that the results are indeed driven by the access to the SC
status: if the access to the SC status is to have an impact on school choices, it can only
affect the cohorts still at school at the time of its implementation. As a result, since
females leave school much earlier than males, cohorts of females born between 1966 and
1971 should be much less affected than their male counterpart.
Why such large differences across genders? Women in India are generally at a disadvantage in terms of schooling. For example, in the sample, women have on average
2.8 years of schooling as opposed to 5.8 for their males counterpart. One of the main
reasons usually put forward to explain this difference are the lower returns to schooling (Kingdon, 1998), or lower perceived returns to schooling (Dreze and Sen, 2002) for
females. Indeed, in the NFHS2 for example, women’s participation in the labour force
is much smaller than males’, with 63% of women “not classified by occupations”24 as
opposed to 14% of males. As a result, women’s returns to education are often perceived
as nonexistent. Moreover, as marriage practices are generally patrilocal, the perception
of the investment in the daughter’s skills is often seen as a waste from the point of view of
the parents. Dreze and Sen (2002) write for example: “[...] the gender division of labour
[...] tends to reduce the perceived benefits of female education. [...] It is in the light of
these social expectations about the adult life of women that female education appears
24
Which essentially means being a housewife.
12
13
7289
-0.03
[0.085]
[-0.16, 0.09]
-0.09
[0.090]
[-0.25, 0.07]
7320
0
[0.101]
[-0.16, 0.16]
0.13
[0.132]
[-0.08, 0.32]
14609
-0.03
[0.082]
[-0.15, 0.09]
0.028
[0.069]
[-0.08, 0.14]
Schooling
7289
-0.085
[0.075]
[-0.21, 0.03]
-0.044
[0.072]
[-0.16, 0.07]
7320
-0.017
[0.092]
[-0.16, 0.12]
0.272**
[0.109]
[0.09, 0.46]
14609
-0.045
[0.067]
[-0.15, 0.06]
0.110*
[0.062]
[0.01, 0.21]
Primary
14609
-0.07
[0.051]
[-0.16, 0.02]
0.165***
[0.051]
[0.07, 0.27]
7320
-0.027
[0.065]
[-0.14, 0.09]
0.245***
[0.069]
[0.11, 0.38]
7289
-0.07
[0.098]
[-0.23, 0.1]
-0.016
[0.090]
[-0.18, 0.14]
7289
-0.150*
[0.082]
[-0.28, -0.02]
0.083
[0.065]
[-0.01, 0.18]
Female Sample
7320
0.057
[0.133]
[-0.15, 0.26]
0.169
[0.160]
[-0.09, 0.41]
Male Sample
14609
-0.036
[0.096]
[-0.19, 0.12]
0.082
[0.083]
[-0.06, 0.21]
Literate
Some Secondary
Full Sample
7289
-0.034
[0.075]
[-0.16, 0.09]
0.026
[0.055]
[-0.06, 0.12]
7320
-0.082
[0.124]
[-0.3, 0.12]
0.140*
[0.075]
[-0.00, 0.29]
14609
-0.056
[0.075]
[-0.19, 0.06]
0.097**
[0.049]
[0.001, 0,2]
Secondary
7289
-0.086
[0.069]
[-0.19, 0.02]
0.065
[0.053]
[-0.02, 0.15]
7320
-0.144*
[0.084]
[-0.29, 0.01]
0.107**
[0.050]
[0.01, 0.20]
14609
-0.087
[0.059]
[-0.19, 0.02]
0.088**
[0.039]
[0.01, 0.17]
Higher Education
*Significant at the 10%, **significant at the 5%, ***significant at the 1%. Standard errors two way clustered at the jati and region level in parenthesis. 95%
confidence interval from two way clustering at the jati and region level, bootstrapped at the region level in brackets. Controls include regions FE, state-cohort
FE, jati FE, gender FE and religion of household head FE. Population born after 1950 and aged 17 and above at the time of the survey.
7289
-1.145
[0.831]
[-2.58, 0.14]
New SC*post 1971
N
0.214
[0.562]
[-0.57, 1.02]
New SC*post 1966
7320
-0.54
[0.640]
[-1.76, 0.66]
New SC*post 1971
N
2.521***
[0.745]
[1.20, 3.78]
New SC*post 1966
14609
-0.873
[0.593]
[-1.98, 0.15]
New SC*post 1971
N
1.473***
[0.399]
[0.64, 2.32]
New SC*post 1966
Years of Schooling
Table 4: Educational attainment: varying treatment year. OLS regressions.
to many parents to be of somewhat uncertain value, if not quite ’pointless’ .”2526 .
The NFHS2 survey contains a question on the reason why a person did not attend
school. Table 5 presents the responses to this question by gender. The main reasons for
males are first and foremost the cost, and then that the child is required for household
work or work on farm or family business. For females, the picture is strikingly different:
the main reasons for which they do not attend school is because it is not considered
necessary, closely followed by the fact that they are required for household work, and
that schooling costs too much. Hence both cost and opportunity costs appear to be a
major reason for which children are not sent to school, with the additional issue that
often, education is not considered necessary for girls, which confirms the qualitative
evidence discussed earlier.
Table 5: Reason not going to school by gender.
main reason never went to school
Female
Male
school too far away
transport not available
education not considered necessary
required for household work
required for work on farm/family business
required for outside work for payment
cost too much
no proper school facilities for girls
required for care of siblings
not interested in studies
other
don’t know
3.38
0.91
23.69
20
3.71
2.6
19.84
3.45
2.49
6.85
5
7.65
2.44
0.76
11.94
12.76
12.47
8.03
25.56
Total
N
100
4 496
100
2 093
0.72
12.47
4.3
8.55
In this context of lower perceived returns from education for girls from the point of
view of the parental household, a gender neutral increase in returns to education and
decrease in costs of education might have different effects by gender. Indeed, education
of the children is costly, both in terms of direct cost and opportunity costs. As a result,
the perception by the parents that returns might be gender-asymmetric either because
25
Dreze and Sen (2002) also underline a second important reason for the low enrolment of women:
the low quality of schooling infrastructures. Indeed while, say, the absence of a nearby school could be
thought to be gender neutral, in practice, girls are more affected, as parents tend to be more reluctant to
send their daughters to far away schools than their sons. Finally, an other reason put forward by Dreze
and Sen (2002) for the lower education of girls is that even if education can be an asset on the marriage
market, it can only be so if the girl’s education remains lower than that of her potential husband.
26
Note that this tendency is in no way specific to the lower castes in India. Actually, lower castes
tend to be less gender biased than higher castes (Srinivas (1966), Chakravarti (1993), Mencher (1988),
Kapadia (1997), Field et al. (2010), Cassan and Van de Walle (2016)).
14
they do not believe that returns to schooling exist for girls, or because they benefit less
from the returns to schooling of girls than that of boys is sufficient a mechanism to drive
those type of behaviour. In addition, under imperfect access to credit and budgetary
constraints, a household may not have the resources to increase the education of both
genders, and favour the ones with highest perceived returns to education. In the most
extreme cases, this may even lead to a reallocation of resources from female to male child,
leading to an actual decrease of education of females. Banerjee and Duflo (2011) discuss
a similar type of reallocation of schooling investment across children. Hence, a plausible
interpretation of the asymmetric results by gender of the access to the affirmative action
program are gender norms.
4
Robustness check
4.1
Differential trend
A concern with any difference in difference estimation are differential trends. A first
check is to examine visually the data. I run the following regression:
Eduidt = constant + βN ewSCid + δ1 N ewSCid ∗ Ct + γ1 Ct
(3)
+ λSit + idt
In which Ct is a vector of 5 years cohorts and Sit a vector of state fixed effects. Figure
1 plots the coefficients on N ewSCid ∗ Ct of these regressions. One can see that there
does not seem to be any major pre-trend driving the results, while both the increase for
males and the decrease for females are clear for treated cohorts.
Let’s now control more formally for pre-trend by adding a “new” SC specific trend
to the specification. I run the following regressions:
Eduidt = constant + βN ewSCid + δN ewSCid ∗ posttreatmentt + γposttreatmentt
ρN ewSCid ∗ trendt + πtrendt + λXidt + idt
(4)
The notation is the same as earlier and trendt stands for a linear time trend. As can
be seen in Table 6, the results on males become larger but less precisely estimated.
For females however, the negative results, already only weakly significant, seem to both
decrease in magnitude and lose significance. As a consequence, the negative result on
female can be interpreted as evidence that women did not benefit from the SC status
15
Figure 1: Cohort specific effects. Coefficients of the interaction terms “New SC”* years cohorts dummies
(controlling for state FE). Population born after 1950 and aged 18 and above at time of survey.
Full sample
-2 -1.5 -1
-.5
0
.5
1
1.5
2
dep. var. years of education
1956-1960
1961-1965
1966-1970
1971-1975
birth cohort
1976-1980
1981-1985
Coefficient on New SC * Cohort
Male sample
-2 -1.5 -1
-.5
0
.5
1
1.5
2
dep. var. years of education
1956-1960
1961-1965
1966-1970
1971-1975
birth cohort
1976-1980
1981-1985
Coefficient on New SC * Cohort
Female sample
-2 -1.5 -1 -.5
0
.5
1 1.5 2
dep. var. years of education
1956-1960
1961-1965
1966-1970
1971-1975
birth cohort
1976-1980
1981-1985
Coefficient on New SC * Cohort
for educational attainment, with only weak evidence that they could have lost from it.
16
17
3.010**
[1.395]
[0.77, 5.16]
-0.134
[0.838]
[-1.74, 1.53]
New SC*post 1966
New SC*post 1971
0.396
[0.709]
[-0.75, 1.55]
-0.975
[0.916]
[-2.75, 0.65]
New SC*post 1966
New SC*post 1971
7289
0.026
[0.097]
[-0.13, 0.18]
-0.027
[0.113]
[-0.24, 0.19]
0.022
[0.098]
[-0.15, 0.19]
-0.022
[0.115]
[-0.24, 0.20]
7320
0.188*
[0.103]
[0.03, 0.34]
0.366**
[0.166]
[0.1, 0.6]
0.171
[0.115]
[-0.13, 0.34]
0.355**
[0.164]
[0.1, 0.59]
Schooling
7289
-0.067
[0.089]
[-0.23, 0.09]
-0.024
[0.096]
[-0.17, 0.13]
-0.072
[0.089]
[-0.24, 0.09]
-0.038
[0.097]
[-0.19, 0.13]
7320
0.014
[0.098]
[-0.16, 0.17]
0.309*
[0.182]
[0.05, 0.57]
0
[0.100]
[-0.18, 0.16]
0.308*
[0.181]
[0.05, 0.57]
Primary
Some Secondary
7320
-0.045
[0.103]
[-0.2, 0.11]
0.221
[0.156]
[-0.02, 0.47]
-0.054
[0.099]
[-0.21, 0.09]
0.224
[0.153]
[-0.02, 0.48]
7289
-0.078
[0.117]
[-0.27, 0.13]
-0.023
[0.117]
[-0.23, 0.19]
-0.082
[0.116]
[-0.28, 0.13]
-0.039
[0.116]
[-0.25, 0.18]
7289
-0.119
[0.100]
[-0.29, 0.04]
0.119*
[0.067]
[0.01, 0.23]
-0.099
[0.097]
[-0.27, 0.60]
0.095
[0.065]
[-0.03, 0.21]
Female Sample
7320
0.174
[0.152]
[-0.07, 0.42]
0.315*
[0.164]
[0.07, 0.57]
0.16
[0.158]
[-0.08, 0.42]
0.305*
[0.167]
[0.04, 0.57]
Male Sample
Literate
7289
-0.055
[0.073]
[-0.2, 0.08]
0.003
[0.063]
[-0.1, 0.11]
-0.055
[0.067]
[-0.18, 0.07]
-0.008
[0.055]
[-0.11, 0.09]
7320
-0.086
[0.150]
[-0.32, 0.15]
0.134
[0.087]
[-0.06, 0.33]
-0.092
[0.147]
[-0.32, 0.14]
0.139
[0.089]
[-0.06, 0.35]
Secondary
7289
-0.107
[0.066]
[-0.22, 0.00]
0.042
[0.060]
[-0.06, 0.01]
-0.100*
[0.060]
[-0.20, 0.01]
0.021
[0.052]
[-0.06, 0.10]
7320
-0.205**
[0.093]
[-0.37, -0.03]
0.029
[0.089]
[-0.14, 0.20]
-0.206**
[0.092]
[-0.37, -0.04]
0.041
[0.094]
[-0.14, 0.22]
Higher Education
*Significant at the 10%, **significant at the 5%, ***significant at the 1%. Standard errors two way clustered at the jati and region level in parenthesis. 95%
confidence interval from two way clustering at the jati and region level, bootstrapped at the region level in brackets. Controls include regions FE, state-cohort
FE, jati FE, gender FE and religion of household head FE. Population born after 1950 and aged 17 and above at the time of the survey.
7289
-0.908
[0.900]
[-2.70, 0.71]
New SC*post 1971
N
0.202
[0.686]
[-0.94, 1.41]
New SC*post 1966
7320
-0.267
[0.806]
[-1.79, 1.38]
New SC*post 1971
N
3.018**
[1.380]
[0.79, 5.16]
New SC*post 1966
Years of Schooling
Table 6: Educational attainment: varying treatment year and including differential trend. OLS regressions.
4.2
Selective migration and identity manipulation
This section will deal with two additional concerns: migration and identity manipulation. Empirically, I attribute the “New SC” status based on the district of residence of
respondents at the time of the survey. Therefore, if prior to 1976, some households have
migrated in order to benefit from the SC status, I would wrongly code them as “Old
SC”, which may bias the results. However, migration is relatively low in India (Munshi
and Rosenzweig, 2009) and particularly so in the 1970’s and earlier. Indeed, from the
Migration Volumes of the 1981 Census, one can estimate that only 6.1% of the males
and 11.1% of females had changed district before 197727 . A similar argument can be
made about identity manipulation (Cassan, 2015): people may well have been incited
to manipulate their caste identity in order to be considered as eligible to the SC status
before 1976. However, as for migration, I do not believe that this behaviour may bias
the results. Indeed, first and foremost, as the data is collected in 1999, 25 years after
the harmonisation of the SC lists, there are actually no incentive to manipulate caste
identity at the time of survey. Second, one may worry that individuals who manipulated their caste identity prior to 1976 permanently changed their jati identity. This
second pattern would result in a miscoding of the treatment status very similar to the
one due to migration: individuals that should have been coded as “New SC” had they
not migrated or manipulated their identity are coded as “Old SC”. To my knowledge,
there is no estimates of the amount of identity manipulation that the SC status may lead
to. However, Cassan (2015) shows that in the context of colonial Punjab, a caste based
policy with arguably stronger impact on day to day life led to an identity manipulation
movement of up to 7.5% of the population, which I’ll use as an upper bound on caste
identity manipulation.
It is to be noted that the households choosing to migrate/manipulate would probably
be the ones that would have benefited the most from the access to the SC status had
they not migrated. This means that, if anything, migration/identity manipulation is
likely to bias the estimates for males downwards and for females upwards, as I would
consider those households as “Old SC”. As the NFHS2 data does not offer useful migration information28 , it is not possible to directly verify if the migrating/manipulating
households indeed are different in any form from the other households. Despite the weak
probability that migration or manipulation could be an issue for my estimation strategy,
27
Note that this is an overestimation of the type of migration that concerns us, since a bias might
emerge if those migrants moved in a way that would change their SC status, which is not systematically
the case when crossing a district border since regions are larger than districts.
28
And obviously no information about identity manipulation.
18
I will show in this subsection that the results hold even with very strong assumptions
on the profiles of migrants/manipulators.
As a robustness check, I reallocate the individuals who may contribute the most to
the results from control to treatment. I identify the individuals that are the most likely to
contribute to bias the results if they had been migrants/manipulators. That is, for male, I
identify “old SC” individuals born prior the first treatment cohort with an above median
occupation, and the ones born after the first treatment cohort with a below median
education (and symmetrically for female). Among these, I randomly draw a number
corresponding to 33%, 66% and 100% of the total migrating/manipulating population29 ,
reallocate them to the “New SC” status and bootstrap 1000 times the main estimations
at each of those steps. Figure 2 plots the coefficients of those estimations. Figure 2 plots
the coefficient on New SC*post1966 (for males) and New SC*post1971 (for females). It
can be seen that the results are very robust, since even under these strong and counter
intuitive assumptions about the education profile of migrants, the male result remains
positive even if 100% of migrants/manipulators had an above the median education, and
remain significant up to 66% of the total population of migrants/manipulators.
5
Conclusion
This paper studies the impact of affirmative action policies for SC in India. It shows
that they have an ambiguous effect on schooling attainment. Indeed, the overall null
effect actually hides differences across genders, with male benefiting largely from those
policies while female to the contrary seem not to be affected, or may even suffer from
it. This suggests that individuals at the intersection of different discriminated groups
may not be sufficiently well protected by public policies. More generally, this paper
underlines the need to focus on the role of accumulation of discrimination, the question
of “intersectionality”, which so far has been neglected in the economics literature.
29
I take the largest population between the estimates of migration and manipulation. That is migrants/manipulators that would account for 7.5% of the total male “New SC” and 11.1% of the total
female “New SC”.
19
-3
-2
-1
0
1
2
3
Figure 2: Reallocation of potential migrants and robustness of the coefficients.
0
33
% re-allocated
66
Coefficient on New SC * Post 1966
99
95% CI
-3
-2
-1
0
1
(a) Male sample, coefficient on New SC*post1966
0
33
% re-allocated
66
Coefficient on New SC * Post 1971
99
95% CI
(b) Female sample, coefficient on New SC*post1971
20
References
Ashraf, Nava, Nathalie Bau, Nathan Nunn, and Alessandra Voena, “Bride
Price and Female Education,” Working Paper, 2015.
Banerjee, Abhijit and Esther Duflo, Poor Economics: A Radical Rethinking of the
Way to Flight Global Poverty, Public Affairs, 2011.
Bayly, Susan, Caste, Society and Politics in India, Cambridge University Press, 1999.
Bertrand, Marianne, Rema Hanna, and Sendhil Mullainathan, “Affirmative
Action in Education: Evidence from College Admissions in India,” Journal of Public
Economics, 2010, 94 (1-2), 16–29.
Cassan, Guilhem, “Identity Based Policies and Identity Manipulation: Evidence from
Colonial Punjab.,” American Economic Journal: Economic Policy, 2015, 7 (4), 103–
31.
and Lore Van de Walle, “The Social Cost of Social Norms: Gender Quotas meet
Gender Norms,” Working Paper, 2016.
Chakravarti, Uma, “Conceptualising Brahmanical Patriarchy in Early India: Gender,
Caste, Class and State,” Economic and Political Weekly, 1993, 28 (14).
Chalam, K.S., “Caste Reservations and Equality of Opportunity in Education,” Economic and Political Weekly, 1990, 25 (41).
Chitnis, Suma, “Education for Equality: Case of Scheduled Castes in Higher Education,” Economic and Political Weekly, 1972, 7 (31-33).
Cuskey, Judson, “cgmwildboot,” Stata ado file, 2015.
Dreze, Jean and Amartya Sen, India. Development and Participation., 2 ed., Oxford
University Press, 2002.
Field, Erica, Rohini Pande, and Seema Jayachandran, “Do Traditional Institutions Constrain Female Entrepreneurship? A Field Experiment on Business Training
in India,” American Economic Review Papers and Proceedings, 2010.
Foster, Andrew and Mark Rosenzweig, “Technological change and the distribution
of schooling: evidence from green-revolution India,” Journal of Development Economics, 2003, 74 (1), 87–111.
21
Galanter, Marc, Competing Equalities: Law and the Backward Classes in India, University of California Press, 1984.
Government of India, ed., Report of the Commissioner for Scheduled Castes and
Scheduled Tribes, 1957-1958., Manager Govt. Of India Press, 1958.
, ed., Report of the Commissioner for Scheduled Castes and Scheduled Tribes, 19581959., Manager Govt. Of India Press, 1960.
, ed., Report of the Commissioner for Scheduled Castes and Scheduled Tribes, 19751977., Manager Govt. Of India Press, 1978.
Hnatkovska, Viktoria, Amartya Lahiri, and Paul Sourabh B., “Castes and Labor
Mobility,” American Economic Journal: Applied Economics, 2012, 4 (2).
,
, and
, “Breaking the Caste Barrier: Intergenerational Mobility in India,”
Journal of Human Resources, 2013, 48 (2).
Hughes, Melanie, “Intersectionality, Quotas, and Minority Women’s Political Representation Woldwide,” American Political Science Review, 2011.
, “Quota Design and the Political Representation of Women from Indigenous and
Tribal Groups,” Working Paper, 2014.
Jensen, Robert, “Do Labor Market Opportunities Affect Young Women’s Work and
Family Decisions? Experimental Evidence from India,” Quarterly Journal of Economics, 2012, 2 (127), 153–192.
Jensenius, Francesca, “Competing Inequalities ? On the Intersection of Gender and
Ethnicity in Candidate Nominations in Indian Elections,” Working Paper, 2015.
Kapadia, Karin, “Mediating the Meaning of Market Opportunities: Gender, Caste
and Class in Rural South India,” Economic and Political Weekly, 1997, 35.
Khanna, Gaurav, “That’s Affirmative: Incentivizing Standards or Standardizing Incentives? Affirmative Action in India,” Working Paper, 2013.
Kingdon, Geetah, “Does the Labour Market Explain Lower Female Schooling in India
?,” Journal of Development Studies, 1998, 35 (1).
Kitts, Eustace J., A Compendium of the Castes and Tribes found in India, Education
Society Press, Byculla, 1885.
22
Krishna, Kala and Veronica Frisancho Robles, “Affirmative Action in Higher
Education in India: Targeting, Catch Up, and Mismatch,” Working Paper, 2012.
Mencher, Joan P, “Women’s Work and Poverty: Women’s Contribution to Household
Maintenance in South India,” in Daisy Dwyer and Judith Bruce, eds., Home Divided,
Stanford University Press, 1988.
Munshi, Kaivan and Mark Rosenzweig, “Why is Mobility in India so Low? Social
Insurance, Inequality, and Growth,” Working Paper, 2009.
Prakash, Nishit, “The Impact of Employment Quotas on the Economic Lives of Disadvantaged Minorities in India,” Working Paper, 2009.
Rosenzweig, Mark and T. Paul Schultz, “Market Opportunities, Genetic Endowments, and Intrafamily Resource Distribution: Child Survival in Rural India,” American Economic Review, 1982, 72 (4), 803–815.
Srinivas, Caste in Modern India and Other Essays, University of California Press, 1966.
Appendices
A
Areas of restriction
The following list presents state by state the different “regions” for which separate SC
list were drawn until 1976. This information has been extracted from the 1971 Census’
Social and Cultural Tables, publicly available, which also contain the whole SC lists by
region. It can be seen that certain “regions” are included in larger ones. In that case,
in the empirical exercises of the paper, I took the most conservative approach, that is:
each region is given a specific fixed effect, while for the computation of the clustered
standard errors, those regions are included in larger ones (except is that would lead to
create only one region for the state). In addition, in certain cases, only taluks are listed,
not entire districts. In that case, I took the most conservative coding approach, which
is to code the whole district as part of the region. In total, the paper uses 31 regions.
Andhra Pradesh
Region A: throughout the state;
23
Region B : districts of Srikakulam, Visakhapathnam, East Godavari, West Godavari, Krishna, Gunter, Ongole, Nellore, Chittoor, Cuddapah, Anantapur and Kurnool;
Region C : districts of Mahbubnagar, Hyderabad, Medak, Nizamabad, Adilabad, Karimnagar, Warangal, Khamman and Nalgonda.
Assam
No regional caste list. This state is not used in the paper.
Bihar
Region A: throughout the state;
Region B : Patna and Tirhut divisions and Monghyr, Bhagalpur, Palamay and Purnea
districts; Region C : districts of Patna, Shahabad, Gaya and Palamau.
Gujarat
Region A: throughout the state except Rajkot division and Kutch district;
Region B : Dangs districts and Umbergaon taluka of Surat district;
Region C : Rajkot division;
Region D: Kutch district.
Haryana
Region A: throughout the state;
Region B : throughout the state except in Mahendragarh and Jind districts;
Region C : districts of Mahendragarh and Jind.
Himachal Pradesh
As Himachal Pradesh is carved out of Punjab in 1966, it is not used in the paper.
Region A: districts of Chamba (excluding the towns of Dalhousie MC, Dalhousie CB and
Bakloh CB), Mandi, Bilaspur, Mahasu, Sirmaur and Kinnaur;
Region B : districts of Kangra, Kulu, Lahul & Spiti, Simla and the towns of Dalhousie
MC, Dalhousie CB and Bakloh CB in Chamba district.
Jammu & Kashmir
No regional caste list. This state is not used in the paper.
Kerala
Region A: Throughout the state;
Region B : throughout the state except in Kasaragod taluka of Malabar district;
Region C : throughout the state except Malabar district (excluding Kasaragod taluka);
Region D: throughout the state except Malabar district;
Region E : Malabar district;
Region F : Malabar district (excluding Kasaragod taluka);
Region G: Kasaragod taluka of Malabar district.
Madhya Pradesh
24
Region A: districts of Bind, Gird, Morena, Shivpuri, Goona, Rajgarh, Shajapur, Ujjain,
Ratlam, Mandsaur, Bhilsa, Indore, Dewas, Dhar, Jhabua and Nimar;
Region B : districts of Chhindwara, Seoni, Betul, Jabalpur, Sagar, Damoh, Mandla,
Hoshangabad, Narsimharpur, Nimar, Balaghat, Raipur, Bilaspur, Durg, Rastar, Surguja and Raigarh;
Region C : Bilaspur district;
Region D: Sagar and Damoh districts;
Region E : Damoh district;
Region F : districts of Bilaspur, Durg, Raipur, Bastar, Surguja and Raigarh;
Region G: Sagar district;
Region H : Balaghat district;
Region I : districts of Balaghat, Bilaspur, Durg, Raipur, Surguja, Bastar and Raigarh;
Region J : districts of Balaghat, Betul, Bilaspur, Durg, Nimar, Raipur, Bastar, Surguja,
Chhindwara, Sagar and Raigarh, Hoshangabad and Seoni Malwa thesils of Hoshangabad
district;
Region K : districts of Sagar and Damoh, Hoshangabad and Seoni Malwa thesils of
Hoshangabad district;
Region L: districts of Chhindwara, Seoni, Betul, Jabalpur, Narsimhapur, Sagar, Damoh,
Mandla, Nimar, Balaghat, Raipur, Durg, Bastar, Surguja, Raigarh and Hoshangabad
(except Harda and Sohagpur thesils);
Region M : districts of Chhindwara, Seoni, Betul, Jabalpur, Narsimhapur, Sagar, Damoh,
Mandla, Nimar, Balaghat, Raipur, Bilaspur, Durg, Bastar, Surguja, Raigarh and Hoshangabad
(except Harda and Sohagpur thesils);
Region N : Sohagpur thesil of Hoshangabad district;
Region O: districts of Datia, Tikamgarh, Chhatarpur, Panna, Stana, Rewa, Sidhi and
Shahdol;
Region P : districts of Raisen and Sehore.
Maharashtra
Region A: throughout the state except Buldana, Akola, Amravati, Yeotmal, Wardha,
Nagpur, Bhandara, Chanda, Aurangabad, Parbhani, Nanded, Bhir, Osmanabad and
Rajura districts;
Region B : districts of Greater Bombay, Dhulia, Jalgaon, Nasik, Ahmednagar, Poona,
Satara, Sangli, Kolhapur, Sholapur, Thana, Kolaba and Ratnagiri;
Region C : districts of Buldana, Akola, Amravati, Yeotmal, Wardha, Nagpur, Bhandara
and Chanda;
Region D: districts of Akola, Amravati and Buldana;
25
Region E : Bhandara district;
Region F : districts of Bhandara and Buldana;
Region G: districts of Amravati, Bhandara and Buldana;
Region H : districts of Aurangabad, Parbhani, Nanded, Rajura, Bhir and Osmanabad.
Manipur
No regional caste list. This state is not used in the paper.
Meghalaya
No regional caste list. This state is not used in the paper.
Karnataka
Region A: throughout the state except Coorg, Belgaum, Bijapur, Dharwar, Kanara,
South Kanara, Gulbarga, Raichur and Bidar districts, and Kollegal talukla of Mysore
district;
Region B : districts of Belgaum, Bijapur, Dharwar and Kanara;
Region D: Kanara district;
Region E : districts of Gulbarga, Bidar and Raichar;
Region F : district of South Kanara and Kollegal taluka of Mysore district;
Region G: Kollegal taluka of Mysore district;
Region H : district of South Kanara;
Region I : Coorg district.
Orissa
Region A: throughout the state;
Region B : Sambalpur district.
Punjab
Even though there were sub regions, those were not removed in 1976, so Punjab is not
used in this paper.
Region A: throughout the state;
Region B : throughout the state except Patiala, Bhatinda, Kapurthala and Sangrur districts;
Region C : districts of Patiala, Bhatinda, Kapurthala and Sangrur districts.
Rajasthan
Region A: throughout the state except Ajmer district, Abu Road taluka of Sirohi district
and Sunel Tappa taluka of Jhalawar district;
Region B : Ajmer district;
Region C : Abu Road taluka of Sirohi district;
Region D: Sunel Tappa taluka of Jhalawar district.
Tamil Nadu
26
Region A: throughout the state;
Region B : throughout the state except Kanyakumari district and Shencottah taluka of
Tirunelveli district;
Region C : Nilgiri district;
Region D: districts of Coimbatore and Salem;
Region E : Kanyakumari district and Shencottah taluka of Tirunelveli district;
Region F : Tanjore district.
Tripura
No regional caste list. This state is not used in the paper.
Uttar Pradesh
Region A: throughout the state; Region B : throughout the state excluding Agra, Meerut
and Rohilkhand divisions;
Region C : Bundelkhand division and the portion of Mirzarpur district south of Kaimur
range.
West Bengal
Region A: throughout the state;
Region B : throughout the state except in Purulia district and the territories transferred
from Purnea district of Bihar;
Region C : in Purulia district and the territories transferred from Purnea district of Bihar;
Region D: the territories transferred from Purnea district of Bihar.
Arunachal Pradesh
No regional caste list. This state is not used in the paper.
Chandigarh
No regional caste list. This state is not used in the paper.
Dadra and Nagar Haveli
No regional caste list. This state is not used in the paper.
Delhi
No regional caste list. This state is not used in the paper.
Goa, Daman and Diu
No regional caste list. This state is not used in the paper.
Mizoram
No regional caste list. This state is not used in the paper.
Pondicherry
No regional caste list. This state is not used in the paper.
27